I didn't catch the sample size. Power was my other thought. If those five categorical variables can be closely approximated by one, then collapsing them is pretty valuable.